src/gallium/auxiliary/vl/vl_idct.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2010 Christian König
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include "vl_idct.h"
  29 #include "vl_vertex_buffers.h"
  30 #include "util/u_draw.h"
  31 #include <assert.h>
  32 #include <pipe/p_context.h>
  33 #include <pipe/p_screen.h>
  34 #include <util/u_inlines.h>
  35 #include <util/u_sampler.h>
  36 #include <util/u_format.h>
  37 #include <tgsi/tgsi_ureg.h>
  38 #include "vl_types.h"
  39
  40 #define BLOCK_WIDTH 8
  41 #define BLOCK_HEIGHT 8
  42
  43 #define SCALE_FACTOR_16_TO_9 (32768.0f / 256.0f)
  44
  45 #define STAGE1_SCALE 4.0f
  46 #define STAGE2_SCALE (SCALE_FACTOR_16_TO_9 / STAGE1_SCALE)
  47
  48 #define NR_RENDER_TARGETS 1
  49
  50 struct vertex_shader_consts
  51 {
  52    struct vertex4f norm;
  53 };
  54
  55 enum VS_INPUT
  56 {
  57    VS_I_RECT,
  58    VS_I_VPOS,
  59
  60    NUM_VS_INPUTS
  61 };
  62
  63 enum VS_OUTPUT
  64 {
  65    VS_O_VPOS,
  66    VS_O_BLOCK,
  67    VS_O_TEX,
  68    VS_O_START
  69 };
  70
  71 static const float const_matrix[8][8] = {
  72    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  73    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  74    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  75    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  76    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  77    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  78    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  79    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  80 };
  81
  82 static void *
  83 create_vert_shader(struct vl_idct *idct, bool calc_src_cords)
  84 {
  85    struct ureg_program *shader;
  86    struct ureg_src scale;
  87    struct ureg_src vrect, vpos;
  88    struct ureg_dst t_vpos;
  89    struct ureg_dst o_vpos, o_block, o_tex, o_start;
  90
  91    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  92    if (!shader)
  93       return NULL;
  94
  95    t_vpos = ureg_DECL_temporary(shader);
  96
  97    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
  98    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
  99
 100    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 101
 102    /*
 103     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 104     *
 105     * t_vpos = vpos + vrect
 106     * o_vpos.xy = t_vpos * scale
 107     * o_vpos.zw = vpos
 108     *
 109     * o_block = vrect
 110     * o_tex = t_pos
 111     * o_start = vpos * scale
 112     *
 113     */
 114    scale = ureg_imm2f(shader,
 115       (float)BLOCK_WIDTH / idct->destination->width0,
 116       (float)BLOCK_HEIGHT / idct->destination->height0);
 117
 118    ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
 119    ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), scale);
 120    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 121    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 122
 123    if(calc_src_cords) {
 124       o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK);
 125       o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX);
 126       o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_START);
 127
 128       ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect);
 129       ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 130       ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, scale);
 131    }
 132
 133    ureg_release_temporary(shader, t_vpos);
 134
 135    ureg_END(shader);
 136
 137    return ureg_create_shader_and_destroy(shader, idct->pipe);
 138 }
 139
 140 static void
 141 fetch_one(struct ureg_program *shader, struct ureg_dst m[2],
 142           struct ureg_src tc, struct ureg_src sampler,
 143           struct ureg_src start, struct ureg_src block, float height)
 144 {
 145    struct ureg_dst t_tc, tmp;
 146    unsigned i, j;
 147
 148    t_tc = ureg_DECL_temporary(shader);
 149    tmp = ureg_DECL_temporary(shader);
 150
 151    m[0] = ureg_DECL_temporary(shader);
 152    m[1] = ureg_DECL_temporary(shader);
 153
 154    /*
 155     * t_tc.x = right_side ? start.x : tc.x
 156     * t_tc.y = right_side ? tc.y : start.y
 157     * m[0..1].xyzw = tex(t_tc++, sampler)
 158     */
 159    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(tc, TGSI_SWIZZLE_X));
 160    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(start, TGSI_SWIZZLE_Y));
 161
 162 #if NR_RENDER_TARGETS == 8
 163    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Z), ureg_scalar(block, TGSI_SWIZZLE_X));
 164 #else
 165    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Z), ureg_imm1f(shader, 0.0f));
 166 #endif
 167
 168    for(i = 0; i < 2; ++i) {
 169       for(j = 0; j < 4; ++j) {
 170          /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */
 171          ureg_TEX(shader, tmp, TGSI_TEXTURE_3D, ureg_src(t_tc), sampler);
 172          ureg_MOV(shader, ureg_writemask(m[i], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X));
 173
 174          if(i != 1 || j != 3) /* skip the last add */
 175             ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y),
 176                ureg_src(t_tc), ureg_imm1f(shader, 1.0f / height));
 177       }
 178    }
 179
 180    ureg_release_temporary(shader, t_tc);
 181    ureg_release_temporary(shader, tmp);
 182 }
 183
 184 static void
 185 fetch_four(struct ureg_program *shader, struct ureg_dst m[2],
 186            struct ureg_src tc, struct ureg_src sampler,
 187            struct ureg_src start, bool right_side, float width)
 188 {
 189    struct ureg_dst t_tc;
 190
 191    t_tc = ureg_DECL_temporary(shader);
 192    m[0] = ureg_DECL_temporary(shader);
 193    m[1] = ureg_DECL_temporary(shader);
 194
 195    /*
 196     * t_tc.x = right_side ? start.x : tc.x
 197     * t_tc.y = right_side ? tc.y : start.y
 198     * m[0..1] = tex(t_tc++, sampler)
 199     */
 200    if(right_side) {
 201       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_Y));
 202       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_X));
 203    } else {
 204       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_X));
 205       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_Y));
 206    }
 207
 208    ureg_TEX(shader, m[0], TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 209    ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_src(t_tc), ureg_imm1f(shader, 1.0f / width));
 210    ureg_TEX(shader, m[1], TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 211
 212    ureg_release_temporary(shader, t_tc);
 213 }
 214
 215 static void
 216 matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2])
 217 {
 218    struct ureg_dst tmp[2];
 219    unsigned i;
 220
 221    for(i = 0; i < 2; ++i) {
 222       tmp[i] = ureg_DECL_temporary(shader);
 223    }
 224
 225    /*
 226     * tmp[0..1] = dot4(m[0][0..1], m[1][0..1])
 227     * dst = tmp[0] + tmp[1]
 228     */
 229    ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0]));
 230    ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(l[1]), ureg_src(r[1]));
 231    ureg_ADD(shader, dst, ureg_src(tmp[0]), ureg_src(tmp[1]));
 232
 233    for(i = 0; i < 2; ++i) {
 234       ureg_release_temporary(shader, tmp[i]);
 235    }
 236 }
 237
 238 static void *
 239 create_transpose_frag_shader(struct vl_idct *idct)
 240 {
 241    struct pipe_resource *transpose = idct->textures.individual.transpose;
 242    struct pipe_resource *intermediate = idct->textures.individual.intermediate;
 243
 244    struct ureg_program *shader;
 245
 246    struct ureg_src block, tex, sampler[2];
 247    struct ureg_src start[2];
 248
 249    struct ureg_dst m[2][2];
 250    struct ureg_dst tmp, fragment;
 251
 252    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 253    if (!shader)
 254       return NULL;
 255
 256    block = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 257    tex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_CONSTANT);
 258
 259    sampler[0] = ureg_DECL_sampler(shader, 0);
 260    sampler[1] = ureg_DECL_sampler(shader, 1);
 261
 262    start[0] = ureg_imm1f(shader, 0.0f);
 263    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 264
 265    fetch_four(shader, m[0], block, sampler[0], start[0], false, transpose->width0);
 266    fetch_one(shader, m[1], tex, sampler[1], start[1], block, intermediate->height0);
 267
 268    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 269
 270    tmp = ureg_DECL_temporary(shader);
 271    matrix_mul(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), m[0], m[1]);
 272    ureg_MUL(shader, fragment, ureg_src(tmp), ureg_imm1f(shader, STAGE2_SCALE));
 273
 274    ureg_release_temporary(shader, tmp);
 275    ureg_release_temporary(shader, m[0][0]);
 276    ureg_release_temporary(shader, m[0][1]);
 277    ureg_release_temporary(shader, m[1][0]);
 278    ureg_release_temporary(shader, m[1][1]);
 279
 280    ureg_END(shader);
 281
 282    return ureg_create_shader_and_destroy(shader, idct->pipe);
 283 }
 284
 285 static void *
 286 create_matrix_frag_shader(struct vl_idct *idct)
 287 {
 288    struct pipe_resource *matrix = idct->textures.individual.matrix;
 289    struct pipe_resource *source = idct->textures.individual.source;
 290
 291    struct ureg_program *shader;
 292
 293    struct ureg_src tc[2], sampler[2];
 294    struct ureg_src start[2];
 295
 296    struct ureg_dst l[2], r[2];
 297    struct ureg_dst t_tc, tmp, fragment[NR_RENDER_TARGETS];
 298
 299    unsigned i;
 300
 301    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 302    if (!shader)
 303       return NULL;
 304
 305    t_tc = ureg_DECL_temporary(shader);
 306    tmp = ureg_DECL_temporary(shader);
 307
 308    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 309    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 310
 311    sampler[0] = ureg_DECL_sampler(shader, 1);
 312    sampler[1] = ureg_DECL_sampler(shader, 0);
 313
 314    start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 315    start[1] = ureg_imm1f(shader, 0.0f);
 316
 317    for (i = 0; i < NR_RENDER_TARGETS; ++i)
 318        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 319
 320    fetch_four(shader, l, tc[0], sampler[0], start[0], false, source->width0);
 321    ureg_MUL(shader, l[0], ureg_src(l[0]), ureg_scalar(ureg_imm1f(shader, STAGE1_SCALE), TGSI_SWIZZLE_X));
 322    ureg_MUL(shader, l[1], ureg_src(l[1]), ureg_scalar(ureg_imm1f(shader, STAGE1_SCALE), TGSI_SWIZZLE_X));
 323
 324    for (i = 0; i < NR_RENDER_TARGETS; ++i) {
 325
 326 #if NR_RENDER_TARGETS == 8
 327       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_imm1f(shader, 1.0f / BLOCK_WIDTH * i));
 328       fetch_four(shader, r, ureg_src(t_tc), sampler[1], start[1], true, matrix->width0);
 329 #elif NR_RENDER_TARGETS == 1
 330       fetch_four(shader, r, tc[1], sampler[1], start[1], true, matrix->width0);
 331 #else
 332 #error invalid number of render targets
 333 #endif
 334
 335       matrix_mul(shader, fragment[i], l, r);
 336       ureg_release_temporary(shader, r[0]);
 337       ureg_release_temporary(shader, r[1]);
 338    }
 339
 340    ureg_release_temporary(shader, t_tc);
 341    ureg_release_temporary(shader, tmp);
 342    ureg_release_temporary(shader, l[0]);
 343    ureg_release_temporary(shader, l[1]);
 344
 345    ureg_END(shader);
 346
 347    return ureg_create_shader_and_destroy(shader, idct->pipe);
 348 }
 349
 350 static void *
 351 create_empty_block_frag_shader(struct vl_idct *idct)
 352 {
 353    struct ureg_program *shader;
 354    struct ureg_dst fragment;
 355
 356    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 357    if (!shader)
 358       return NULL;
 359
 360    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 361
 362    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 363
 364    ureg_END(shader);
 365
 366    return ureg_create_shader_and_destroy(shader, idct->pipe);
 367 }
 368
 369 static bool
 370 init_shaders(struct vl_idct *idct)
 371 {
 372    idct->matrix_vs = create_vert_shader(idct, true);
 373    idct->matrix_fs = create_matrix_frag_shader(idct);
 374
 375    idct->transpose_vs = create_vert_shader(idct, true);
 376    idct->transpose_fs = create_transpose_frag_shader(idct);
 377
 378    idct->eb_vs = create_vert_shader(idct, false);
 379    idct->eb_fs = create_empty_block_frag_shader(idct);
 380
 381    return
 382       idct->transpose_vs != NULL && idct->transpose_fs != NULL &&
 383       idct->matrix_vs != NULL && idct->matrix_fs != NULL &&
 384       idct->eb_vs != NULL && idct->eb_fs != NULL;
 385 }
 386
 387 static void
 388 cleanup_shaders(struct vl_idct *idct)
 389 {
 390    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 391    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 392
 393    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 394    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 395
 396    idct->pipe->delete_vs_state(idct->pipe, idct->eb_vs);
 397    idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs);
 398 }
 399
 400 static bool
 401 init_buffers(struct vl_idct *idct)
 402 {
 403    struct pipe_resource template;
 404    struct pipe_sampler_view sampler_view;
 405    struct pipe_vertex_element vertex_elems[2];
 406    unsigned i;
 407
 408    memset(&template, 0, sizeof(struct pipe_resource));
 409    template.last_level = 0;
 410    template.depth0 = 1;
 411    template.bind = PIPE_BIND_SAMPLER_VIEW;
 412    template.flags = 0;
 413
 414    template.target = PIPE_TEXTURE_2D;
 415    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 416    template.width0 = idct->destination->width0 / 4;
 417    template.height0 = idct->destination->height0;
 418    template.depth0 = 1;
 419    template.usage = PIPE_USAGE_STREAM;
 420    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 421
 422    template.target = PIPE_TEXTURE_3D;
 423    template.format = PIPE_FORMAT_R16_SNORM;
 424    template.width0 = idct->destination->width0 / NR_RENDER_TARGETS;
 425    template.depth0 = NR_RENDER_TARGETS;
 426    template.usage = PIPE_USAGE_STATIC;
 427    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 428
 429    for (i = 0; i < 4; ++i) {
 430       if(idct->textures.all[i] == NULL)
 431          return false; /* a texture failed to allocate */
 432
 433       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 434       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 435    }
 436
 437    idct->vertex_bufs.individual.quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks);
 438
 439    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 440       return false;
 441
 442    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 443    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 444    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 445    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 446    (
 447       idct->pipe->screen,
 448       PIPE_BIND_VERTEX_BUFFER,
 449       sizeof(struct vertex2f) * 4 * idct->max_blocks
 450    );
 451
 452    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 453       return false;
 454
 455    /* Rect element */
 456    vertex_elems[0].src_offset = 0;
 457    vertex_elems[0].instance_divisor = 0;
 458    vertex_elems[0].vertex_buffer_index = 0;
 459    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 460
 461    /* Pos element */
 462    vertex_elems[1].src_offset = 0;
 463    vertex_elems[1].instance_divisor = 0;
 464    vertex_elems[1].vertex_buffer_index = 1;
 465    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 466
 467    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 468
 469    return true;
 470 }
 471
 472 static void
 473 cleanup_buffers(struct vl_idct *idct)
 474 {
 475    unsigned i;
 476
 477    assert(idct);
 478
 479    for (i = 0; i < 4; ++i) {
 480       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 481       pipe_resource_reference(&idct->textures.all[i], NULL);
 482    }
 483
 484    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 485    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 486    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 487 }
 488
 489 static void
 490 init_state(struct vl_idct *idct)
 491 {
 492    struct pipe_sampler_state sampler;
 493    unsigned i;
 494
 495    idct->viewport[0].scale[0] = idct->textures.individual.intermediate->width0;
 496    idct->viewport[0].scale[1] = idct->textures.individual.intermediate->height0;
 497
 498    idct->viewport[1].scale[0] = idct->destination->width0;
 499    idct->viewport[1].scale[1] = idct->destination->height0;
 500
 501    idct->fb_state[0].width = idct->textures.individual.intermediate->width0;
 502    idct->fb_state[0].height = idct->textures.individual.intermediate->height0;
 503
 504    idct->fb_state[0].nr_cbufs = NR_RENDER_TARGETS;
 505    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 506       idct->fb_state[0].cbufs[i] = idct->pipe->screen->get_tex_surface(
 507          idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, i,
 508          PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 509    }
 510
 511    idct->fb_state[1].width = idct->destination->width0;
 512    idct->fb_state[1].height = idct->destination->height0;
 513
 514    idct->fb_state[1].nr_cbufs = 1;
 515    idct->fb_state[1].cbufs[0] = idct->pipe->screen->get_tex_surface(
 516       idct->pipe->screen, idct->destination, 0, 0, 0,
 517       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 518
 519    for(i = 0; i < 2; ++i) {
 520       idct->viewport[i].scale[2] = 1;
 521       idct->viewport[i].scale[3] = 1;
 522       idct->viewport[i].translate[0] = 0;
 523       idct->viewport[i].translate[1] = 0;
 524       idct->viewport[i].translate[2] = 0;
 525       idct->viewport[i].translate[3] = 0;
 526
 527       idct->fb_state[i].zsbuf = NULL;
 528    }
 529
 530    for (i = 0; i < 4; ++i) {
 531       memset(&sampler, 0, sizeof(sampler));
 532       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 533       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 534       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 535       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 536       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 537       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 538       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 539       sampler.compare_func = PIPE_FUNC_ALWAYS;
 540       sampler.normalized_coords = 1;
 541       /*sampler.shadow_ambient = ; */
 542       /*sampler.lod_bias = ; */
 543       sampler.min_lod = 0;
 544       /*sampler.max_lod = ; */
 545       /*sampler.border_color[0] = ; */
 546       /*sampler.max_anisotropy = ; */
 547       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 548    }
 549 }
 550
 551 static void
 552 cleanup_state(struct vl_idct *idct)
 553 {
 554    unsigned i;
 555
 556    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 557       idct->pipe->screen->tex_surface_destroy(idct->fb_state[0].cbufs[i]);
 558    }
 559
 560    idct->pipe->screen->tex_surface_destroy(idct->fb_state[1].cbufs[0]);
 561
 562    for (i = 0; i < 4; ++i)
 563       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 564 }
 565
 566 struct pipe_resource *
 567 vl_idct_upload_matrix(struct pipe_context *pipe)
 568 {
 569    struct pipe_resource template, *matrix;
 570    struct pipe_transfer *buf_transfer;
 571    unsigned i, j, pitch;
 572    float *f;
 573
 574    struct pipe_box rect =
 575    {
 576       0, 0, 0,
 577       BLOCK_WIDTH,
 578       BLOCK_HEIGHT,
 579       1
 580    };
 581
 582    memset(&template, 0, sizeof(struct pipe_resource));
 583    template.target = PIPE_TEXTURE_2D;
 584    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 585    template.last_level = 0;
 586    template.width0 = 2;
 587    template.height0 = 8;
 588    template.depth0 = 1;
 589    template.usage = PIPE_USAGE_IMMUTABLE;
 590    template.bind = PIPE_BIND_SAMPLER_VIEW;
 591    template.flags = 0;
 592
 593    matrix = pipe->screen->resource_create(pipe->screen, &template);
 594
 595    /* matrix */
 596    buf_transfer = pipe->get_transfer
 597    (
 598       pipe, matrix,
 599       u_subresource(0, 0),
 600       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 601       &rect
 602    );
 603    pitch = buf_transfer->stride / sizeof(float);
 604
 605    f = pipe->transfer_map(pipe, buf_transfer);
 606    for(i = 0; i < BLOCK_HEIGHT; ++i)
 607       for(j = 0; j < BLOCK_WIDTH; ++j)
 608          f[i * pitch + j] = const_matrix[j][i]; // transpose
 609
 610    pipe->transfer_unmap(pipe, buf_transfer);
 611    pipe->transfer_destroy(pipe, buf_transfer);
 612
 613    return matrix;
 614 }
 615
 616 static void
 617 xfer_buffers_map(struct vl_idct *idct)
 618 {
 619    struct pipe_box rect =
 620    {
 621       0, 0, 0,
 622       idct->textures.individual.source->width0,
 623       idct->textures.individual.source->height0,
 624       1
 625    };
 626
 627    idct->tex_transfer = idct->pipe->get_transfer
 628    (
 629       idct->pipe, idct->textures.individual.source,
 630       u_subresource(0, 0),
 631       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 632       &rect
 633    );
 634
 635    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 636 }
 637
 638 static void
 639 xfer_buffers_unmap(struct vl_idct *idct)
 640 {
 641    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 642    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 643 }
 644
 645 bool
 646 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix)
 647 {
 648    assert(idct && pipe && dst);
 649
 650    idct->pipe = pipe;
 651    pipe_resource_reference(&idct->textures.individual.matrix, matrix);
 652    pipe_resource_reference(&idct->textures.individual.transpose, matrix);
 653    pipe_resource_reference(&idct->destination, dst);
 654
 655    idct->max_blocks =
 656       align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
 657       align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
 658       idct->destination->depth0;
 659
 660    if(!init_buffers(idct))
 661       return false;
 662
 663    if(!init_shaders(idct)) {
 664       cleanup_buffers(idct);
 665       return false;
 666    }
 667
 668    if(!vl_vb_init(&idct->blocks, idct->max_blocks)) {
 669       cleanup_shaders(idct);
 670       cleanup_buffers(idct);
 671       return false;
 672    }
 673
 674    if(!vl_vb_init(&idct->empty_blocks, idct->max_blocks)) {
 675       vl_vb_cleanup(&idct->blocks);
 676       cleanup_shaders(idct);
 677       cleanup_buffers(idct);
 678       return false;
 679    }
 680
 681    init_state(idct);
 682
 683    xfer_buffers_map(idct);
 684
 685    return true;
 686 }
 687
 688 void
 689 vl_idct_cleanup(struct vl_idct *idct)
 690 {
 691    vl_vb_cleanup(&idct->blocks);
 692    vl_vb_cleanup(&idct->empty_blocks);
 693    cleanup_shaders(idct);
 694    cleanup_buffers(idct);
 695
 696    cleanup_state(idct);
 697
 698    pipe_resource_reference(&idct->destination, NULL);
 699 }
 700
 701 void
 702 vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 703 {
 704    unsigned tex_pitch;
 705    short *texels;
 706
 707    unsigned i;
 708
 709    assert(idct);
 710
 711    if(block) {
 712       tex_pitch = idct->tex_transfer->stride / sizeof(short);
 713       texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 714
 715       for (i = 0; i < BLOCK_HEIGHT; ++i)
 716          memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 717
 718       vl_vb_add_block(&idct->blocks, false, x, y);
 719    } else {
 720
 721       vl_vb_add_block(&idct->empty_blocks, true, x, y);
 722    }
 723 }
 724
 725 void
 726 vl_idct_flush(struct vl_idct *idct)
 727 {
 728    struct pipe_transfer *vec_transfer;
 729    struct quadf *vectors;
 730    unsigned num_blocks, num_empty_blocks;
 731
 732    assert(idct);
 733
 734    vectors = pipe_buffer_map
 735    (
 736       idct->pipe,
 737       idct->vertex_bufs.individual.pos.buffer,
 738       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 739       &vec_transfer
 740    );
 741
 742    num_blocks = vl_vb_upload(&idct->blocks, vectors);
 743    num_empty_blocks = vl_vb_upload(&idct->empty_blocks, vectors + num_blocks);
 744
 745    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, vec_transfer);
 746
 747    xfer_buffers_unmap(idct);
 748
 749    if(num_blocks > 0) {
 750
 751       /* first stage */
 752       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[0]);
 753       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[0]);
 754
 755       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 756       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 757       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]);
 758       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 759       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 760       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 761
 762       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_blocks * 4);
 763
 764       /* second stage */
 765       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 766       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 767
 768       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 769       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 770       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[1]);
 771       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 772       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 773       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 774
 775       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_blocks * 4);
 776    }
 777
 778    if(num_empty_blocks > 0) {
 779
 780       /* empty block handling */
 781       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 782       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 783
 784       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 785       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 786       idct->pipe->bind_vs_state(idct->pipe, idct->eb_vs);
 787       idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs);
 788
 789       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, num_blocks * 4, num_empty_blocks * 4);
 790    }
 791
 792    xfer_buffers_map(idct);
 793 }